Google Is Launching AI Chips Into Space — Project Suncatcher's First Satellite Lifts Off October 1

Google is putting its AI hardware where no TPU has gone before: orbit. On September 24, the company announced that the first satellite for Project Suncatcher — a refrigerator-sized prototype named MVP, built with Planet Labs — will launch on October 1 aboard SpaceX's Transporter-18 rideshare mission on a Falcon 9 out of Vandenberg Space Force Base in California. The payload is the interesting part. MVP carries four of Google's Trillium tensor processing units — the same custom AI chips that power its Gemini models — fed by about one kilowatt of solar panels. That is roughly what a single data-center server draws, and Google is explicit about what this mission is and isn't: nobody is training Gemini in orbit. The chips will run short Gemini workloads — roughly 15-minute windows before shutting down to cool — while engineers collect data on whether commercial AI chips can survive launch forces, radiation, and the thermal extremes of space. The ground testing so far is promising. Google replicated launch vibration through three-axis vibration testing (individual components can face 50 to 100 times Earth's gravity), and ran the Trillium TPUs under a proton beam at UC Davis's Crocker Nuclear Laboratory while running live AI workloads. The hardware tolerated a total ionizing dose beyond what a five-year orbital mission would deliver, and most radiation-induced errors could be cleared by simply restarting the chips. Cooling is the other big engineering puzzle. In a vacuum there is no air to carry heat away, so Google built a system of heat pipes and radiators, validated inside a thermal vacuum chamber. In 2027 the company plans two more satellites to test the high-bandwidth laser links needed to connect computing satellites into clusters — designs under study envision 81-satellite constellations linked by 1.6 Tbps optical interconnects in a dawn-dusk orbit. Why it matters: AI's appetite for electricity is the industry's biggest bottleneck, and orbit has one overwhelming advantage — solar panels at orbital altitudes receive up to eight times more energy annually than on the ground, with near-perpetual sunlight in dawn-dusk orbits. Google says this first mission is "about seeing what works, identifying points of failure, and applying those findings to future missions." Analysts caution that cooling limits and launch costs put commercial orbital data centers years away — but the October 1 data will be decisive. If the chips come back healthy, the solar-powered AI data center stops being a research paper and becomes an engineering problem with a deadline.